Advanced AI Programming

A herd of Lumala making use of pathfinding, flocking behavior, and local avoidance to flee from the player as a group.


Role

AI Lead on Kindred Fates, architecting and implementing comprehensive NPC AI systems for an open-world action combat RPG.


Key Achievements

  • Architected hybrid AI system combining Hierarchical Finite State Machines with focused Behavior Trees, reducing graph complexity by 80%
  • Delivered sophisticated combat AI supporting 50+ unique NPC combat AIs for action combat with distinct behaviors and tactical capabilities
  • Implemented multi-layered architecture enabling modular, testable, and maintainable AI systems.
  • Created advanced stealth mechanics using terrain detail maps for seamless tall grass detection without additional setup
  • Created custom tooling for efficiently authoring new combat AI, setting up patrol paths, and debugging npc behavior

Architecture & Systems

Advanced Navigation & Pathfinding

  • Implemented world-spanning Recast navigation mesh with 128x128m tiling for efficient runtime modifications
  • Designed asynchronous pathfinding system preventing frame rate impact during navigation calculations
  • Built sophisticated local avoidance algorithms enabling natural crowd dynamics during large-scale battles

Hybrid Decision-Making Framework

  • Refactored from monolithic behavior trees to HFSM + Behavior Tree hybrid, eliminating cyclic dependencies
  • HFSMs manage state transitions and high-level decisions; Behavior Trees handle discrete action sequences
  • Integrated custom Hierarchical Task Network (HTN) system for dynamic combat planning and goal decomposition

Intelligent Combat Systems

  • Developed centralized WorldState system eliminating data duplication and ensuring consistent NPC decision-making
  • Created comprehensive debugging tools for real-time AI state visualization and behavior analysis
  • Implemented social aggro system enabling coordinated multi-NPC encounters and tactical group behaviors

Impact

Transformed AI development from unwieldy, hard-to-debug systems into clean, maintainable architecture supporting complex gameplay scenarios. The modular design enabled rapid iteration on NPC behaviors while maintaining code quality, directly supporting the team's ability to create diverse and challenging encounters across the open world.